Greetings!

I'm trying to perform data analysis obtained from Agilent 4x44 Arabidopsis two-color microarray using R.

All the samples were made with 3 repeatings. The r- and gDyeNormSignal were used for the analysis (intra-array normalized values obtained from Agilent Feature Extraction Software). The next analysis was performed using "limma" package for R. Then the Aquantile between-array normalisation was conducted. The design matrix defines substraction of a wild type organism arrays from mutant ones.

And the second variant - one of the reference mutants values were substracted from other ones. Empyrical Bayes statistics was performed.

The differential expressed genes were extracted using benjamini-hochberg correction method.

Then it suddenly appears that one of the samples has all it's adjusted p.values above 0.26! The fact is - it is when substracting the mutant values. With the reference of wild type it's alright.

If I understand the process correctly, the adjusted p.values must mean, whether the difference between same sample repeatings are significant or not (please, correct me if I am wrong). And the correction is used because first we need to compare 4 points oof one chip, and second - different chips. So why is it different depending on what reference I use?

What should I do to resolve it? Should I change the adjust method to less strict?